Re: implementation note for scapegoat tree

On Mar 28, 3:10 pm, Ben Pfaff <b...@xxxxxxxxxxxxxxx> wrote:
"user923005" <dcor...@xxxxxxxxx> writes:
On Mar 28, 10:51 am, Ben Pfaff <b...@xxxxxxxxxxxxxxx> wrote:
[scapegoat tree implementation description]
How well does your tree perform?

I don't have a performance testing framework set up, just a
correctness testing framework. Someday I need to revisit the
performance testing I did for my libavl paper and find out.
Until then, I'm pleased with the code.

If you want to look at it, here's the source:
http://cvs.savannah.gnu.org/viewcvs/pspp/src/libpspp/Attic/bt.c?rev=1...
http://cvs.savannah.gnu.org/viewcvs/pspp/src/libpspp/Attic/bt.h?rev=1...
http://cvs.savannah.gnu.org/viewcvs/pspp/tests/libpspp/Attic/bt-test....
The dependencies on other parts of that source tree are intended
to be minimal.

You're still working on PSPP?
Here are some things I would find interesting (if they are not already
present):
Generic Levenberg Marquardt fitting of families of functions to trend
data (specifically, polynomial degree N, exponential(N), log(N),
N*log(N), 2^N, N!, N^N) and linear combinations thereof.
It would be very nice to examine a population and see what population
model fits it best (e.g., automatically identify the distribution as
Gaussian, poisson, etc.)
I would like to see the following intervals calculated:
Confidence intervals
Tolerance intervals
Prediction intervals.

Here is something that would be stupid-cool. Take a big database
vector and look at it. Tell the user what kind of a population it
represents. Fit a curve to the data, and show the three kinds
(confidence, tolerance, and prediction) that surround the data.

Imagine (for instance) that you are buying stocks. You could draw
curves that fit the data and know (with a given level of confidence)
that the price was going up.

Imagine (for instance) that you manufacture bolts. You could sample
the population a few times and know how many bolts per batch were not
going to meet specs.

If PSPP can already do these things I will be very interested to know
it.

.